In Japanese patent unexamined publication JP2001-157057A, the relationship between the amount of the luminance noise N and the signal level D which is converted to the density value is expressed with a function, N=abcD. Here, a, b, and c are constant terms and are statically provided. However, since the amount of the luminance noise changes dynamically due to reasons such as the temperature, the exposure duration, and the gain during the shooting, the above method that uses static constant terms cannot deal with expressing the relationship with a function by adapting with the amount of noise originated during the shooting, and it is therefore inferior in the accuracy of the presumption of the amount of noise. Moreover, although the frequency characteristics of the filtering is controlled using the amount of noise, since this filtering processes the flat region and the edge region in the image equivalently without distinguishing from each other, the edge region in an area that is presumed as having a large amount of noise from the signal level is deteriorated, and therefore, the preservation of the original signal was not highly achieved. Furthermore, it cannot deal with color noise originated between the color signals.
In Japanese patent un-examined publication JP2001-175843A, an input signal is separated into a luminance signal and a color difference signal, edge intensity is obtained based on those signals, and smoothing processing is performed to the color difference signal in flat areas other than the edge region, and wherein this smoothing processing is fixedly performed regardless of the signal level. However, since the amount of the color noise differs with the signal level, the smoothing processing cannot be optimally controlled, and it is possible that the color noise component remains or the deterioration of the original signal occurs, for example.
In order to deal with this problem, JP2006-023959A proposes a method that obtains the similarities of the hues, divides into areas, and applies a noise model for each area, during the noise coring processing based on the noise models. Additionally, JP2006-101006A proposes a method that obtains a threshold for the coring based on the noise model using the average value of a pixel of interest and its neighborhood in order to presume the luminance level, performs a segmentation of the image again in accordance with the similarity of the pixel within the threshold, and applies a noise model for each area.